Bayesian ANN-Based Prediction and Multi-Objective Optimization of Tribological Behavior in Magnesium Alloy AZ91D at Elevated Temperature Using Pareto GA

This study uses an artificial neural network (ANN) model to predict the wear rate and friction coefficient of the magnesium alloy AZ91D, based on experimental data from a pin-on-disc tribometer. The model includes three essential process parameters: sliding velocity (m/s), applied load (kg), and sliding distance (km), in addition to the chamber temperature (°C). A total of 27 experimental designs were devised using a Box-Behnken design. The ANN model was trained utilizing the Bayesian regularization approach with one hidden layer of 10 neurons. The developed ANN models for predicting wear rate and coefficient of friction were used as goal functions in a multi-objective Pareto-based genetic algorithm to maximize tribological performance. The ideal solution indicates a sliding velocity of 2 m/s, a load of 5 kg, a sliding distance of 1.5 km, and a chamber temperature of 143°C, yielding a minimal wear rate of 1.7891 mm3/kg·km and a coefficient of friction of 0.1435. Energy Dispersive Spectroscopy and Scanning Electron Microscopy analyses of worn surfaces show that the wear rate decreases with increasing load and sliding velocity at higher temperatures. The oxide layer that forms at high temperatures enhances wear resistance, even under high loads and sliding speeds.